Planning for astroid 3.0

[cdce8p]

With the breaking change for the new Try node we should probably start planning for version 3.0. We might as well use the major version bump to bundle other breaking changes as well.

How to update pylint

With the tight coupling between pylint and astroid, one of the main challenges will be how to keep the changes to pylint manageable. After each major change, we could

• Release a new alpha version of astroid
• Or pin a specific astroid commit in pylint directly.

astroid @ git+https://github.com/cdce8p/astroid.git@<commit>

Either way would help us keep the pylint changes reviewable.

Any bugfix release would only be merged in the pylint maintenance branch.

ToDo's

• New Try node
  Add Try node #1867
• Redo Import nodes to better align with current AST -> Add Alias node
  Add ImportAlias node as children of Import nodes #912
  Turn ImportFrom.names into a list of NamedTuple #1339
• Fully remove deprecated nodes -> Ellipsis, Index, ExtSlice
  Remove deprecated nodes #2152
• Remove deprecated default argument values. Mostly None for arguments which should never be None. That will help simplify typing.
• Remove deprecated doc attributes
  Remove deprecated doc attribute #2154
• Documentation for all breaking changes -> upgrade guide

Maybe?

• Better typing for Uninferable. Just an idea, with a breaking change we could change the class to UninferableType and Uninferable being an instance of it.
  Consider creating a UninferableType or _Uninferable class #1680
• Make all arguments to __init__ and postinit for nodes keyword only args.
• Remove explicit node exports from astroid.__init__.
• Remove astroid.scoped_nodes module (not the package)

--
Those are just some things I had in mind. Happy to add additional things to it.
/CC: @Pierre-Sassoulas, @DanielNoord, @jacobtylerwalls

Open Questions

Until we're done with all changes, I don't think it make sense to release a new minor version. Should we release 2.13 before we start the process just to get the last features out there? Any bug fixes could be backported as usual. The breaking changes are mostly for downstream libraries, not astroid itself.

[Pierre-Sassoulas]

Should we release 2.13 before we start the process just to get the last features out there? Any bug fixes could be backported as usual. The breaking changes are mostly for downstream libraries, not astroid itself.

Agree with that, the 2.12 maintenance was very long already. I think we waited for #1189 but it can be merged when it's ready. (Edit: #1189 seems to be ready for review, and already 2 persons approved, we should probably add this one in the TODO list.)

[DanielNoord]

TODO list seems good to me. I'd like to help work on stuff, but I think we should try and do some concentrated efforts as that avoid the risks of many PRs ending up in review limbo.

[cdce8p]

Should we release 2.13 before we start the process just to get the last features out there? Any bug fixes could be backported as usual. The breaking changes are mostly for downstream libraries, not astroid itself.

Agree with that, the 2.12 maintenance was very long already. I think we waited for #1189 but it can be merged when it's ready. (Edit: #1189 seems to be ready for review, and already 2 persons approved, we should probably add this one in the TODO list.)

I'll try to take a look at it soon ™️ If it's good, it could make sense to include it with 2.13 before we start with merging 3.0 PRs.

TODO list seems good to me. I'd like to help work on stuff, but I think we should try and do some concentrated efforts as that avoid the risks of many PRs ending up in review limbo.

Agreed, I opened the issue since I wanted to start the discussion early (and had the Try change ready). Ideally, we don't wait too long for the 3.0 release after we start merging breaking changes.

Fortunately, most of the changes are strait forward and easy to implement. The most challenging one left is probably the Import node rewrite. My idea was to prepare a PR beforehand, similarly to the one I already opened for Try before moving forward.

Since the overlap between the changes is minimal, that can be done for some of the other ones too, although not strictly necessary.

I'd like to help work on stuff

These ones only require some time

• Removal of Ellipsis, Index, ExtSlice
• Remove deprecated doc attributes
• Remove explicit node exports from astroid.__init__ (if we want to do it)

There is also the separate docs page, i.e. some kind of upgrade guide. Laying the foundation for it would help (create a file, add link to it from the main page, etc.). After that, everyone can start contributing to it much more easily.

[DanielNoord]

There is also the separate docs page, i.e. some kind of upgrade guide. Laying the foundation for it would help (create a file, add link to it from the main page, etc.). After that, everyone can start contributing to it much more easily.

#1873 😄

Least I could do!

[Pierre-Sassoulas]

I just released 2.12.13, I think it should be the last patch version based on 2.12. I'm wondering if we should release 2.13.0 without #1189. Marc reviewed and there's a lot of comments. We can always release 2.14.0 fast if the situation evolve.

[jacobtylerwalls]

The downside is that it's blocking several pylint PRs, and the unresolved comments are almost entirely cosmetic/small.

[DanielNoord]

If @cdce8p is interested in this I would be happy to do some cooperation on that PR and implement his comments in a back-and-forth process to just merge this somewhere this week.

I have some free time in GMT+1 evening this week so if that would work for you I think we can help move that PR along.

[cdce8p]

If @cdce8p is interested in this I would be happy to do some cooperation on that PR and implement his comments in a back-and-forth process to just merge this somewhere this week.

I have some free time in GMT+1 evening this week so if that would work for you I think we can help move that PR along.

I will try to make some time, although I can't promise it.

--
As for the general timeframe for 3.0. There are still several big todos I would like to take a look at before moving forward. So I think probably at least a month, maybe around New Year.

[jacobtylerwalls]

Duplicate of #2148, the two issues already linked.